coherent: fix cache flush logic if coherent is not the first member

[kv2019i]

If coherent object is at non-zero offset, the init call ends up invalidating data cache outside the allocated buffer.

All current users of "struct coherent" have the object as the first member, so nothing is broken with current code.

[lyakh]

yeah, fixing the same assumption in other locations like

sof/src/include/sof/coherent.h

Line 191 in da9344a

dcache_invalidate_region(uncache_to_cache(c), size);

sof/src/include/sof/coherent.h

Line 217 in da9344a

dcache_invalidate_region(cc, size);

sof/src/include/sof/coherent.h

Line 239 in da9344a

dcache_writeback_invalidate_region(c, size);

(and more) would require some (internal) API changes

[lgirdwood]

@kv2019i @lyakh I assume we need followups for

1. coherent zalloc() alignment to cache line size ?
2. remove the offset, i..e always at offset 0 ?

[lyakh]

@kv2019i @lyakh I assume we need followups for

1. coherent zalloc() alignment to cache line size ?

2. remove the offset, i..e always at offset 0 ?

@lgirdwood I don't think we need those because we so far aren't aware of any bugs that those changes would fix. In fact I think we could add support for the coherent header anywhere within the managed object, apart from type-casting convenience I don't see any reason why it has to be in the beginning.

[kv2019i]

SOFCI TEST

[kv2019i]

A few failes in https://sof-ci.01.org/sofpr/PR7299/build4829/devicetest/index.html (PM fail, also seen on other PRs) an dhttps://sof-ci.01.org/sofpr/PR7299/build4830/devicetest/index.html . In the latter, once case of #7191 was hit (https://sof-ci.01.org/sofpr/PR7299/build4830/devicetest/index.html?model=TGLU_RVP_NOCODEC_IPC4ZPH&testcase=multiple-pause-resume-5 ). Neither related to this PR, so proceeding with merge.